Building a Mind for Machines: Zhu Songchun, BIGAI's CUV Architecture, and the General-Purpose Intelligent Agent Tongtong

At the 2026 World Artificial Intelligence Conference, Zhu Songchun reflected on the large-model approach and argued for replacing pure data fitting with BIGAI's CUV cognitive architecture, promoting the idea of "building a mind for machines." His team has released version 3.0 of its general-purpose intelligent agent Tongtong, upgrading its spatial, cognitive, and social intelligence, and it ranked first overall in eight categories of the team's self-developed General Social AI (GSA) test. This approach still faces three questions: evaluation independence, simulation-to-reality transfer, and a closed commercialization loop.